Revisiting Forgotten Fire Tests: Causal Inference and Counterfactuals for Learning Idealized Fire-Induced Response of RC Columns

The expensive nature and unique facilities required for fire testing make it difficult to conduct comprehensive experimental campaigns. As such, engineers can often afford to test a small number of specimens. This complicates attaining a proper inference, especially when addressing questions in the form of what would have been the fire response of a particular specimen had it been twice as large? Or had it been made from a different material grade? In hindsight, answering causal and hypothetical (counterfactual) questions goes beyond the capacity of statistical and machine learning methods which were built to address observational data. To overcome the above challenges, this paper presents a causal approach to answering such questions. In this approach, principles of causal inference are adopted to reconstruct the deformation-time history of reinforced concrete (RC) columns and propose an idealized fire response for these columns. The findings of this study indicate the significant influence of the loading level, aggregate type, and longitudinal steel ratio on the deformation history of fire-exposed RC columns.

Automated summary

The expensive and unique facilities required for fire testing make it difficult to conduct comprehensive experimental campaigns, resulting in limited testing of specimens. Addressing causal and hypothetical questions about fire response becomes challenging for statistical and machine learning methods. To overcome this, this paper presents a causal approach to answer such questions by adopting principles of causal inference to reconstruct the deformation-time history of reinforced concrete (RC) columns and propose an idealized fire response. The findings highlight the significant influence of loading level, aggregate type, and longitudinal steel ratio on the deformation history of fire-exposed RC columns.